Targeting Local Benchmarks: A Framework for NLP-Based Automated Grading in Bangladesh’s NCTB English Curriculum

Abstract

This study examines the methodological congruence between international standards and Natural Language Processing (NLP)-based assessment within Bangladesh's National Curriculum and Textbook Board (NCTB) English curriculum. Although AI-based grading systems are increasingly used in education, most remain built on Western language norms and tend to overlook local educational contexts and the influence of learners' first language (L1: Bangla). Using a qualitative research design, this study conducted content analysis of 200 student essays (100 at the SSC level and 100 at the HSC level) alongside semi-structured interviews with 15 local education professionals, including curriculum developers, teachers, and examiners. The findings reveal a high degree of contextual mismatch: existing NLP systems have no models calibrated to the NCTB level and tend to over-penalize non-standard but communicatively effective English. The absence of a locally digitized corpus further complicates the development of fair automated grading systems. By shifting the evaluative focus from native-speaker norms to locally competent communicators, this study underscores the importance of incorporating cultural and linguistic context into AI-based assessment tools. It proposes a tentative policy framework to guide policymakers toward culturally responsive NLP systems that support fairer student assessment in Bangladesh.